Unfortunately not, the C2 ranking formula is not yet available in the workbench, so it is not possible to test it. As soon as something changes, I will immediately let you know.
At the moment, I am working on my formula based on the available stats (max open loss, max dd and so on) for backtesting in the workbench. I think it will be interesting; I plan to publish it by the end of February.
Changed two strategies, new ones:
I plotted Cumulative Profit chart for convenience (several times a day I save the Cumulative P/L value).
SmartPortfolio replaced one strategy
By the way, new strategy with skin in the game (TOS).
Strategy Save for Retirement LR dropped out of the C2Star program, it trades mainly long side and the recent correction contributed to the excess of the maximum drawdown limit.
@Daniil Congratulations on this thread. It’s a great exercise and I must confess that I am super curious about the results. I will certainly follow the thread.
I must confess that I have never tried such an exercise of combining a portfolio of several strategies, and at first sight I have 2 concerns, that I would like to know if you thought about it and what were your conclusions.
1- The attrition rate in the C2Star program is very high. From your initial lineup only 2 out of 5 strategies are still in the C2Star program. When you conceptualized this did you took this fact into account?
2- When a strategy has a DD, and every strategy has it, it might take more than 2 weeks to recover. I would be concerned that the reevaluation of lineup every 2 weeks will not give enough time to an individual strategy to recover. If things align in the wrong side of things, the portfolio might be accumulating only the “bad” moments of all strategies and not give them time to recover. Do you think this might happen? What’s your view about this?
Thanks for your interest!
I performed backtests every 15 days, hence the results consider high dropout rates.
When I started working on this, I expected a high dropout rate, but whether it would be profitable I did not know.
It was intriguing to collect data, see how strategies are constantly being replaced, and then insert them into the model and see that it works anyway.
It turns out that you can do without looking for an excellent strategy that will be stable for years (which is a rather utopian task) and focus on the ranking method.
C2Star is essentially a ranking method.
A logical conclusion, I also thought so at first.
But if it increases the rebalance time to 30 days, the results worse. This applies not only to C2Star, but also to the ranking method I am developing.
It turns out that periodically good strategies “go crazy”, basically it happens on an increasing basis, and in this case, a 15-day rebalance is the only thing that prevents this “crazy” strategy from causing severe damage to your portfolio.
But when I was doing this backtest, it was impossible to calculate the use of Stop Losses for the SmartPortfolio in the Workbench, but now this has already been implemented (I added it for my rank method and this improved the result)
After your question I want to recalculate all data of C2Star with stop, and then a 30-day rebalance. I will publish the results here as soon as it is ready.
I’m all talking in this thread about developing my own rank method.
Sorry for the delay, there is already a finished version.
It remains only to add one parameter to the SmartPortfolio settings and I will publish it
what happens to performance if you try to enforce a rule that all strategies in your model portfolio have roughly equal VAR, or some other measure of tail risk? I dont think that at these short horizons returns should even be in the ranking, but rather equal risk would maybe dominate. and with high-variance programs that tend to blow up’, you want a large number of strategies in your portfolio, in as many different asset classes as possible. ranking them by performance at this time scale isnt going to be remotely accurate
Greetings to this thread @GeorgeColes
I am an adept of practice and try not to get involved in theoretical discussions.
“VAR, or some other measure of tail risk” is just about this. The market is about tails, and trading success is the ability to wait for a positive tail while surviving negative ones.
And modeling of tail risks will in no way answer the question: “Does it work or not?”
It can only answer on “Can this work or not, according to the given theoretical model?”
My empirical experience suggests that there is no point in wasting time on perfecting the theoretical model, more productive is move to practice.
There is a C2Star program and the SmartPortfolio backtest based on it showed good results, which means I’m moved to real time testing.
Do not think my answer is rude, this is just my opinion.
As for short horizons (rebalance periods) @JoaquimFonseca
I did new backtests with rebalance period of 15 days and 30 days. Here is a comparison
Stop = false (without any stops)
Stop = OpenLoss 5%
The performance practically does not differ, 15 days a little better (0.9% in absolute value).
Although the presence and absence of a stop in this test did not affect the historical result, I will add it(stop) to the SmartPortfolio setup, since anything can happen with strategies.
thanks for the reply. Calling VAR, a simple empirical statistic of the return distribution, theoretical, tells me all I need to know!
It seems you misunderstood me.
I meant that the VAR has no predictive value and only gives the illusion of the expected risk. This is what I mean by theoretical = theoretical risk.
The market is about extremes, and very often those that are not yet in the accumulated statistics.
So I built the dispersion of return for you, where you can see the tails. And what to do next with this?
I am really bothered by the lack of a good amount of data for this experiment. For that reason alone I don’t like the C2Star concept. I really don’t know how to test short runs of data like this except some kind of resampling approach, and you can at least try to find the bounds of the joint distritubion of returns, but all of it is going to be distorted by the short history across too few market regimes.
I would personally just try to over diversify, normalize my exposure to all the subject programs to something like 1/n or equal marginal risk contribution, and ride it out for a few years. I would abandon the C2Star ranking method and see what kind of portfolios of C2 strategies work best over the longest possible period of time, and dead strategies need to be incorporated to prevent survivorship bias.
I think maybe a good thing to try, since you probably have a lot of data on failed / abandoned strategies, is to perhaps try to predict which strategies are doomed, malformed, or otherwise should be excluded from the master portfolio. I think C2Star is an attempt to pre-screen strategies in this manner, but I think it maybe misses the mark? It seems like maybe “C2Star” should be a backwards-looking simulation, and the “C2 meta-portfolio” is an index of strategies that have a long track record and good risk adjusted performance. The idea that programs that have only been around for a few months are safe to invest in? I don’t know.